Image processing for crop/weed discrimination in fields with high weed pressure

Nan Li, Tony E. Grift, Ting Yuan, Chunlong Zhang, Md Abdul Momin, Wei Li

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Machine vision based automatic crop or weed detection in commercial fields is still an open problem, because of wide variety of plant species and the unstructured lighting condition. The task becomes even more challenging when weeds are prominent and overlap with crops. This paper presents an image processing method for fast discrimination of crops and weeds in fields with high weed infestation levels. Mahalanobis distance was used to classify crop and weed pixels in field images based on their differences in hue and saturation. A Naïve Bayes classifier was built to compare with the Mahalanobis distance based classifier. 80 images (100 by 100 pixels) of celery cabbage, broccoli and weeds were used to train and evaluate the method. Evaluation result showed that this method correctly discriminated 68.0% of crop pixels and 83.2% weed pixels in celery cabbage and weed images, and 97.0% of crop pixels and 99.7% of weed pixels in broccoli and weed images. The average time requirement for processing each 100-by-100-pixel image was 9.7 ms. Compared with the Naïve Bayes classifier, the Mahalanobis distance based classifier was more suitable to address the problem of this study. In addition, this method was built into a crop detection method designed in our previous work. A series of 15 field images of broccoli with high weed pressure were used to test the combined method. The results indicated that the combined method correctly detected 93.6% of the cops, a significant improvement over the previous method.

Original languageEnglish (US)
Title of host publication2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016
PublisherAmerican Society of Agricultural and Biological Engineers
ISBN (Electronic)9781510828759
DOIs
StatePublished - Jan 1 2016
Event2016 ASABE Annual International Meeting - Orlando, United States
Duration: Jul 17 2016Jul 20 2016

Publication series

Name2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016

Other

Other2016 ASABE Annual International Meeting
CountryUnited States
CityOrlando
Period7/17/167/20/16

Fingerprint

Crops
Image processing
weeds
Pixels
image analysis
crops
Classifiers
broccoli
celery
cabbage
methodology
Computer vision
Lighting
computer vision
processing technology
lighting
Processing

Keywords

  • Crop detection
  • Image processing
  • Machine vision
  • Mahalanobis distance
  • Weed control

ASJC Scopus subject areas

  • Bioengineering
  • Agronomy and Crop Science

Cite this

Li, N., Grift, T. E., Yuan, T., Zhang, C., Momin, M. A., & Li, W. (2016). Image processing for crop/weed discrimination in fields with high weed pressure. In 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016 (2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016). American Society of Agricultural and Biological Engineers. https://doi.org/10.13031/aim.20162460475

Image processing for crop/weed discrimination in fields with high weed pressure. / Li, Nan; Grift, Tony E.; Yuan, Ting; Zhang, Chunlong; Momin, Md Abdul; Li, Wei.

2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016. American Society of Agricultural and Biological Engineers, 2016. (2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Li, N, Grift, TE, Yuan, T, Zhang, C, Momin, MA & Li, W 2016, Image processing for crop/weed discrimination in fields with high weed pressure. in 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016. 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016, American Society of Agricultural and Biological Engineers, 2016 ASABE Annual International Meeting, Orlando, United States, 7/17/16. https://doi.org/10.13031/aim.20162460475
Li N, Grift TE, Yuan T, Zhang C, Momin MA, Li W. Image processing for crop/weed discrimination in fields with high weed pressure. In 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016. American Society of Agricultural and Biological Engineers. 2016. (2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016). https://doi.org/10.13031/aim.20162460475
Li, Nan ; Grift, Tony E. ; Yuan, Ting ; Zhang, Chunlong ; Momin, Md Abdul ; Li, Wei. / Image processing for crop/weed discrimination in fields with high weed pressure. 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016. American Society of Agricultural and Biological Engineers, 2016. (2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016).
@inproceedings{b1d2d8fc5cb8414cb18195d3fd60f0db,
title = "Image processing for crop/weed discrimination in fields with high weed pressure",
abstract = "Machine vision based automatic crop or weed detection in commercial fields is still an open problem, because of wide variety of plant species and the unstructured lighting condition. The task becomes even more challenging when weeds are prominent and overlap with crops. This paper presents an image processing method for fast discrimination of crops and weeds in fields with high weed infestation levels. Mahalanobis distance was used to classify crop and weed pixels in field images based on their differences in hue and saturation. A Na{\"i}ve Bayes classifier was built to compare with the Mahalanobis distance based classifier. 80 images (100 by 100 pixels) of celery cabbage, broccoli and weeds were used to train and evaluate the method. Evaluation result showed that this method correctly discriminated 68.0{\%} of crop pixels and 83.2{\%} weed pixels in celery cabbage and weed images, and 97.0{\%} of crop pixels and 99.7{\%} of weed pixels in broccoli and weed images. The average time requirement for processing each 100-by-100-pixel image was 9.7 ms. Compared with the Na{\"i}ve Bayes classifier, the Mahalanobis distance based classifier was more suitable to address the problem of this study. In addition, this method was built into a crop detection method designed in our previous work. A series of 15 field images of broccoli with high weed pressure were used to test the combined method. The results indicated that the combined method correctly detected 93.6{\%} of the cops, a significant improvement over the previous method.",
keywords = "Crop detection, Image processing, Machine vision, Mahalanobis distance, Weed control",
author = "Nan Li and Grift, {Tony E.} and Ting Yuan and Chunlong Zhang and Momin, {Md Abdul} and Wei Li",
year = "2016",
month = "1",
day = "1",
doi = "10.13031/aim.20162460475",
language = "English (US)",
series = "2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016",
publisher = "American Society of Agricultural and Biological Engineers",
booktitle = "2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016",

}

TY - GEN

T1 - Image processing for crop/weed discrimination in fields with high weed pressure

AU - Li, Nan

AU - Grift, Tony E.

AU - Yuan, Ting

AU - Zhang, Chunlong

AU - Momin, Md Abdul

AU - Li, Wei

PY - 2016/1/1

Y1 - 2016/1/1

N2 - Machine vision based automatic crop or weed detection in commercial fields is still an open problem, because of wide variety of plant species and the unstructured lighting condition. The task becomes even more challenging when weeds are prominent and overlap with crops. This paper presents an image processing method for fast discrimination of crops and weeds in fields with high weed infestation levels. Mahalanobis distance was used to classify crop and weed pixels in field images based on their differences in hue and saturation. A Naïve Bayes classifier was built to compare with the Mahalanobis distance based classifier. 80 images (100 by 100 pixels) of celery cabbage, broccoli and weeds were used to train and evaluate the method. Evaluation result showed that this method correctly discriminated 68.0% of crop pixels and 83.2% weed pixels in celery cabbage and weed images, and 97.0% of crop pixels and 99.7% of weed pixels in broccoli and weed images. The average time requirement for processing each 100-by-100-pixel image was 9.7 ms. Compared with the Naïve Bayes classifier, the Mahalanobis distance based classifier was more suitable to address the problem of this study. In addition, this method was built into a crop detection method designed in our previous work. A series of 15 field images of broccoli with high weed pressure were used to test the combined method. The results indicated that the combined method correctly detected 93.6% of the cops, a significant improvement over the previous method.

AB - Machine vision based automatic crop or weed detection in commercial fields is still an open problem, because of wide variety of plant species and the unstructured lighting condition. The task becomes even more challenging when weeds are prominent and overlap with crops. This paper presents an image processing method for fast discrimination of crops and weeds in fields with high weed infestation levels. Mahalanobis distance was used to classify crop and weed pixels in field images based on their differences in hue and saturation. A Naïve Bayes classifier was built to compare with the Mahalanobis distance based classifier. 80 images (100 by 100 pixels) of celery cabbage, broccoli and weeds were used to train and evaluate the method. Evaluation result showed that this method correctly discriminated 68.0% of crop pixels and 83.2% weed pixels in celery cabbage and weed images, and 97.0% of crop pixels and 99.7% of weed pixels in broccoli and weed images. The average time requirement for processing each 100-by-100-pixel image was 9.7 ms. Compared with the Naïve Bayes classifier, the Mahalanobis distance based classifier was more suitable to address the problem of this study. In addition, this method was built into a crop detection method designed in our previous work. A series of 15 field images of broccoli with high weed pressure were used to test the combined method. The results indicated that the combined method correctly detected 93.6% of the cops, a significant improvement over the previous method.

KW - Crop detection

KW - Image processing

KW - Machine vision

KW - Mahalanobis distance

KW - Weed control

UR - http://www.scopus.com/inward/record.url?scp=85009088995&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85009088995&partnerID=8YFLogxK

U2 - 10.13031/aim.20162460475

DO - 10.13031/aim.20162460475

M3 - Conference contribution

AN - SCOPUS:85009088995

T3 - 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016

BT - 2016 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2016

PB - American Society of Agricultural and Biological Engineers

ER -